HodgeNet: Learning Spectral Geometry on Triangle Meshes
arXiv:2104.12826
Abstract
Constrained by the limitations of learning toolkits engineered for other applications, such as those in image processing, many mesh-based learning algorithms employ data flows that would be atypical from the perspective of conventional geometry processing. As an alternative, we present a technique for learning from meshes built from standard geometry processing modules and operations. We show that low-order eigenvalue/eigenvector computation from operators parameterized using discrete exterior calculus is amenable to efficient approximate backpropagation, yielding spectral per-element or per-mesh features with similar formulas to classical descriptors like the heat/wave kernel signatures. Our model uses few parameters, generalizes to high-resolution meshes, and exhibits performance and time complexity on par with past work.
arXiv commentAccepted to SIGGRAPH 2021
References in corpus (10)
- Decoupled Weight Decay Regularization
- Geometric deep learning: going beyond Euclidean data
- Spectral Networks and Locally Connected Networks on Graphs
- Deep Convolutional Networks on Graph-Structured Data
- MeshCNN: A Network with an Edge
- Gauge Equivariant Mesh CNNs: Anisotropic convolutions on geometric graphs
- ZerNet: Convolutional Neural Networks on Arbitrary Surfaces via Zernike Local Tangent Space Estimation
- Parallel Transport Convolution: A New Tool for Convolutional Neural Networks on Manifolds
- MeshWalker: Deep Mesh Understanding by Random Walks
- Weakly Supervised Deep Functional Map for Shape Matching